EnSEMBLE: Enhancer-Set Enrichment from Routine RNA-seq Illuminates Cell-State Dynamics

Summary: Cell-state transitions, rather than static identities, underlie development, plasticity and chemoresistance, yet they can be obscured by conventional gene-set enrichment analysis (GSEA) because gene expression is buffered by post-transcriptional regulation. Crucially, the same bulk or single-cell RNA-seq experiment also captures short, bidirectionally transcribed enhancer RNAs (eRNAs), which report real-time enhancer activity and therefore provide a direct read-out of regulatory state—without any additional sequencing. We present EnSEMBLE—Enhancer-Set Enrichment Method with Biological ExpLanation, a framework that augments GSEA with quantitative enhancer-level evidence. First, context-specific enhancer sets are assembled from gold-standard ATAC-seq and ChIP-seq atlases covering diverse tissues, cell types and transcription-factor binding sites. Second, a rapid analytic rank statistic evaluates whether the eRNAs for each set are preferentially up- or down-stream in paired expression profiles, yielding a normalized enrichment score and exact P-value in milliseconds. Third, a language-model agent integrates conventional pathway enrichment (gene level) with EnSEMBLE scores (enhancer level) to produce plain-language mechanistic summaries, allowing users to triangulate cell-state changes with their upstream regulatory drivers. We applied EnSEMBLE to BT-20 triple-negative breast-cancer cells engineered to over-express BCL2L14-ETV6 fusion variants. The method pinpointed strong activation of enhancer programs characteristic of normal mammary epithelium, fibroblasts and mesenchymal stem cells—an epigenomic signature consistent with epithelial-to-mesenchymal transition, acquisition of stem-like traits and the paclitaxel resistance observed in ETV6-rearranged TNBC. By leveraging eRNA signals already embedded in routine RNA-seq, EnSEMBLE supplies fast, statistically rigorous and biologically interpretable insight into dynamic cell states, providing a drop-in complement to gene-centric analyses for perturbation screens, disease stratification and precision oncology. Akshat Gupta, Renu Sharma, Bashir Lawal, Yue Wang, and Xiaosong Wang